Post-translational modification-centric base editor screens to assess phosphorylation site functionality in high-throughput

Abstract

Signaling pathways that drive gene expression are typically depicted as having a dozen or so landmark phosphorylation and transcriptional events. In reality, thousands of dynamic post-translational modifications (PTMs) orchestrate nearly every cellular function, and we lack technologies to find causal links between these vast biochemical pathways and genetic circuits at scale. Here, we describe the high throughput, functional assessment of phosphorylation sites through the development of PTM-centric base editing coupled to phenotypic screens, directed by temporally-resolved phosphoproteomics. Using T cell activation as a model, we observe hundreds of unstudied phosphorylation sites that modulate NFAT transcriptional activity. We identify the phosphorylation-mediated nuclear localization of PHLPP1 which promotes NFAT but inhibits NFκB activity. We also find that specific phosphosite mutants can alter gene expression in subtle yet distinct patterns, demonstrating the potential for fine-tuning transcriptional responses. Overall, base editor screening of PTM sites provides a powerful platform to dissect PTM function within signaling pathways.

Introduction

Nearly every eukaryotic cellular process is controlled by post-translational modifications (PTMs), which can modulate protein subcellular localization, protein-biomolecular interactions, enzymatic activity, stability, etc. Protein phosphorylation is arguably the best characterized PTM¹. The human genome encodes for roughly 500 protein kinases and 200 phosphatases that control the coupling and hydrolysis of phosphates, respectively, on substrates in a rapid and dynamic fashion^2,3. These signaling cascades organize into elaborate biochemical networks that allow the cell to process information about its intra- and extracellular environmental changes. Mass spectrometry-based proteomics has revolutionized our ability to map global signaling pathways at the phosphorylation site (phosphosite)-specific level, across time and cellular space. Current phosphoproteomics experiments can quantify tens of thousands of phosphosites, tracking their dynamics upon cell stimulation, drug treatment, or mutational status^4–7. Unfortunately, fewer than 3% of the nearly quarter million phosphorylation sites identified have an ascribed function^8,9. Since functionally characterizing novel modification sites is laborious and resource intensive, we often resort to describing complex biological systems by the limited number of well-characterized phosphosites for which we have good reagents.

Functional genomics has greatly increased our throughput for associating genes with specific cellular phenotypes. Genome-wide CRISPR/Cas9 technology coupled with phenotypic screens allow researchers to identify which genes or non-coding regions are important for a specific function such as gene expression¹⁰, cytokine secretion¹¹, cell proliferation¹², or cell survival^13,14. More recently, CRISPR/Cas9-mediated base editors, which introduce specific nucleotide substitutions in genomic DNA rather than double stranded DNA breaks¹⁵, have been used for mutational scanning across protein coding genes and regulatory elements^16–19. Base editor technology holds immense promise to study PTM site function²⁰ in high-throughput by mutating specific amino acids, bypassing the need to create site-specific homology-directed repair templates.

Here, we describe an experimental workflow to study phosphorylation site functionality in high-throughput. By coupling quantitative phosphoproteomics with “proteome-wide” base editing of individual phosphosites and phenotypic screens, we are able to functionally evaluate a large number of previously unstudied phosphosites that are involved in cell proliferation or the transcriptional responses following T cell activation. T cell activation via stimulation of the T cell receptor (TCR) and co-stimulation of CD28 activates multiple kinases that form both negative and positive feedback loops, as well as several transcription factors including NFAT, NFκB, and AP1. Applying PTM-centric base editor screens to T cell activation, we show that we can recapitulate many known aspects of the pathway, while discovering novel kinase activities and specific phosphorylation events that control different aspects of the transcriptional response. This allowed us to identify a specific phosphosite on the phosphatase PHLPP1 as a novel regulator of T cell activation-induced NFAT and NFκB activities. Transcriptional profiling of PHLPP1 phosphosite mutant T cells shows that individual phosphorylation events differentially impact downstream gene expression in subtle yet distinct patterns, creating the potential to map causal links between signaling and gene expression. PTM-centric base editor screens provide an experimental framework to functionally interrogate and systematically decode the vast network of biochemical signaling events to their downstream phenotypes.

Results

Optimization of base editing for experimentally-derived phosphorylation sites

To develop a system by which we could profile and then functionally assess signaling pathways and their effects on gene expression, we focused on a classic model of T cell activation in the human T cell leukemia line Jurkat E6–1, during which multiple kinases and downstream transcription factors are activated (Figure 1A)²¹. We performed a temporally-resolved quantitative phosphoproteomic experiment, assaying global phosphorylation patterns for 0, 3, 9 and 27 minutes of T cell activation using α-CD3 and α-CD28 agonist antibodies^22,23. Of the 26,037 quantified phosphopeptides, 899 were significantly differentially regulated (moderated F test, FDR < 0.05) during this time series (Figure 1B, Supp Table 1, and Supp Data File 1). Replicates showed strong correlation and that the three-to-nine-minute transition showed the largest changes according to principle component analysis (Extended Data Figure 1). PTM-SEA²⁴, a PTM site-centric analog to GSEA²⁵ (gene set enrichment analysis), showed that various kinase activities were temporally regulated during the first 30 minutes of T cell activation (Figure 1C). This analysis culminated with the perturbation signatures for “anti-CD3” and “phorbol esters”, indicating the temporally-regulated phosphopeptides reflected the appropriate T cell activation pathways.

Figure 1.

Signaling dynamics of early T cell activation. A) Diagram of early T cell activation and where the time points for the phosphoproteomic analysis were sampled. Pink ovals represent kinases, colored circles are transcription factors. B) Heatmap of statistically regulated (moderated F test) phosphopeptides during the first 27 minutes of T cell stimulation. Interactive data is associated with this figure (Supp Data File 1). C) Heatmap of PTM-SEA terms for the phosphoproteomics time series data. “iKiP” indicates the In Vitro Kinase-to-Phosphosite Database⁶⁷, “pKS” is kinase-substrate pairs from PhosphositePlus (PSP). “PERT” is a PSP-curated perturbation.

Using a custom bioinformatics pipeline, we queried which of the ~19,000 confidently localized phosphosites we could target using SpCas9-mediated C-to-T or A-to-G editors. We included all detected phosphosites, rather than only temporally regulated ones since there are likely to be phosphosites that were not statistically significant but still contribute to T cell activation. Considering editing windows and targetable locations, we found 7,618 unique phosphosites were targetable with 11,392 distinct sgRNAs using the SpCas9-A-to-G editor ABE8e, while 7,063 unique phosphosites could be targeted by the SpCas9-C-to-T editor BE4 (Figure 2A and Supp Table 2)^26,27. Roughly half of the editable phosphosites overlapped between the two base editors. The amino acid side chain representation of targetable phosphosites reflected those detected and statistically regulated (Figure 2B). ABE8e appears to make more structurally conservative missense mutations, and unlike BE4, can target tyrosine-encoding codons (Figure 2A).

Figure 2.

Base editing capabilities of empirically derived phosphorylation sites. A) The bar graph shows the number of sgRNAs that target phosphosites for the respective base editors BE4 (pink) or ABE8e (blue) and what the resultant amino acid codon will be for BE4 or ABE8e. BE4 cannot mutate Y to any other amino acid and was omitted. The Venn diagram inset shows the number of distinct phosphosites targeted by at least one sgRNA. B) The distribution of all putative phosphorylated amino acid side chains from all phosphopeptides detected by mass spectrometry, the number of statistically regulated phosphopeptides in Figure 1B, and the number of sgRNAs that can be used by ABE8e to target a phosphosite. C) Diagram for testing various base editor delivery methods in an arrayed format, one sgRNA at a time followed by T cell activation. D) Base editing efficiency, as determined by NGS amplicon sequencing in percent of edited reads, testing the nucleofection of different biomolecules to deliver base editors. The base editing window, counting right to left from the PAM sequence, is shown in the inset and the nucleic acid targets are color coordinated. “p” indicates plasmid, “r” is recombinantly expressed and purified, “mRNA” is synthetic, capped mRNA. E) Effect of phosphosite base edits on T cell activation-induced CD69 surface levels. “Edits” indicates the targeted gene, “Stim” indicates 12 hours of treatment with α-CD3/CD28 agonist antibodies, and “Target” indicates amino acid targeted. NTC, non-targeting control at the HEK3 locus; MFI, mean fluorescence intensity. Two sample T test p-values are shown where “ns” denotes not significant. n = two or three editing replicates, indicated by the data points where standard deviation is shown.

To develop a flexible genomic engineering approach, we based our base editing strategy on previous genome-wide CRISPR/Cas9 screens in primary T cells where sgRNAs are delivered via lentivirus, followed by electroporation of Cas9 protein¹². We nucleofected several different types of biomolecules to determine the most efficient base editors (Figure 2C). Using Jurkat cells stably expressing a sgRNA targeting the model HEK3 site in humans, we nucleofected either plasmid DNA, chemically synthesized and capped mRNA, or recombinant protein of different base editor versions. We found that purified recombinant ABE8e protein (NGG PAM,) properly edited over 95% of adenosines in the base editing window (Figure 2D). To test the reproducibility of the ABE8e protein, we re-expressed and purified the protein²⁷. Again, we found that 92% of the adenosines in the base editing window were mutated to guanosine via Sanger sequencing²⁸. These results demonstrate that we can reproducibly achieve sufficiently high base editing efficiency with ABE8e protein for high-throughput screens.

Base editing phosphosites that promote or inhibit markers of T cell activation

We tested whether mutating phosphosites in proteins known to be involved in the TCR signaling pathway with ABE8e protein would affect markers of T cell activation. We base edited the activating tyrosine of MAPK1 (also known as ERK2) Y187, and two targets in the TCR associated kinase ZAP70 Y292 and Y315 in Jurkat E6.1 cells. MAPK1 Y187 acts as a positive control as it is important for proper T cell activation signaling and transcriptional responses²⁹. Mutant or control cells were activated with α-CD3/CD28 agonist antibodies for 12 hours, stained for the early activation marker surface CD69 levels and analyzed by flow cytometry. Mutation of an inhibitory phosphotyrosine on ZAP70 Y292H³⁰ increased CD69 surface expression whereas MAPK1 Y187C showed diminished surface CD69 (Figure 2E and Extended Data Fig. 2). ZAP70 Y315H showed no effect, consistent with previous reports³¹. Together these data establish that phosphosite mutations can have positive or negative effects on T cell activation levels.

Phosphosite-centric functional phenotypic screens using pooled base editing for cell proliferation

To assess base editing efficiency in pooled format, we created a lentiviral library consisting of roughly 11,000 phosphosite-targeting sgRNAs for missense mutations, 250 non-targeting controls, and 250 intergenic controls as negative controls. We also included 250 guides that introduce terminating edits in essential genes via mRNA splice site disruption, effectively knocking out the gene. TPR (triple parameter reporter) Jurkat cells, which have individual fluorescent reporters driven by separate NFAT, NFκB, and AP1 transcriptional response elements³², were transduced at a multiplicity of infection of 0.3. After puromycin selection, a 500x library coverage aliquot of cells was collected and the rest were electroporated with ABE8e protein. To confirm our base editing was efficient we analyzed the representation of sgRNAs, comparing pre- and six days post-ABE8e protein electroporation (Figure 3A). The representation of sgRNAs disrupting splice junctions in 250 essential genes was significantly lower six days post-base editing compared to all sgRNAs in the library (Figure 3B) indicating our base editing approach was working efficiently. The relative representation of intergenic and non-targeting controls was not affected by introduction of ABE8e protein.

Figure 3.

Base editing screening reveals phosphosites involved in proliferation or survival. A) Diagram of pooled base editor screens for phosphosites or terminating edits of essential genes important for cell proliferation or survival followed by next-generation sequencing (NGS). B) Log2 fold change between pre- and post-ABE8e protein introduction of cells expressing the sgRNA library. “Essential” refers to a terminating edit into essential genes. Intergenic base edits and non-targeting controls are also shown. Two sample T test p values are shown, n= 4 transduction replicates. C) Mutations made to CDK1 phosphosites and their influence of sgRNA representation before and after base editing in a pooled format. “sg1 and sg2” indicate that two different sgRNAs were used for the Y15C mutation. Two sample T-test p values are shown, n= 4 transduction replicates where standard deviation is shown. Box and whiskers plot shows median, quartiles, max and min, and outliers (individual data points). D) Volcano plot showing the distribution of all sgRNAs pre- and post-ABE8e protein introduction as determined by MAGeCK analysis. Green points indicate terminating edits in essential genes. Red points indicate statistically significant sgRNA targeting phosphosites depleted after base editing. Blue points are sgRNAs targeting phosphosites that were enriched after base editing. Enrichment values and statistical thresholds were determined by MAGeCK¹³. E) Kinase Library, site-centric enrichment analysis of phosphosite mutants, as an aggregate motif, enriched in post ABE8e-edited cells (blue) or pre ABE8e-edited cells (red) bins. Enrichment Values by MAGeCK¹³ and the one-sided Fisher’s exact text p value, after Benjamini-Hochberg, correction is show.

We next examined whether mutation of phosphosites important for cell division would affect cell viability or proliferation in our pooled format without specific stimuli or selective pressure³³. Examination of phosphorylation sites on CDK1, a kinase whose activity is necessary for proper cell division, showed that Y160C;T161A and Y15C reduced cell viability post-ABE8e electroporation similar in magnitude and direction to levels previously seen using homology-directed recombination³⁴ (Figure 3C). The silent mutation of Y19Y showed no effect. Analysis of the whole phosphosite-mutant dataset using the Model-based Analysis of Genome-wide CRISPR/Cas9 Knockout (MAGeCK)¹³ resulted in 89 sgRNAs that were significantly enriched pre-ABE compared to post-editing (Figure 3D and Supp Table 3). sgRNAs targeting phosphosites that were depleted after ABE8e protein introduction were enriched for genes involved in the term “Cell Cycle” amongst the top three Reactome pathways (Extended Data Figure 3). 58 sgRNAs introducing phosphosite mutations were enriched post-base editing, suggesting a proliferative advantage (Figure 3D). These genes belonged to the Reactome signatures for “membrane trafficking”, “metabolism of RNA” and “transcriptional regulation by RUNX1”, pathways involved in the proliferation of glioblastoma and acute myeloid leukemia cells^35–37 (Extended Data Figure 3). GO analysis of the genes knocked out through splice site disruption showed an enrichment for mRNA processing and cell cycle (Extended Data Figure 3). PTM site-centric pathway analyses can identify groups of phosphosites (in this case, mutation of phosphosites) enriched in the pre- or post-edited pools. Kinase Library³⁸, which uses primary sequence motifs derived from biochemical kinase reactions to predict kinase activity through motif enrichment, identified CDK1/4/5/6/13/18 motifs from aggregated phosphosite mutants depleted relative to pre-edited cells (Figure 3E). MELK1, NIM1, DYRK2/4, and YANK2/3 motifs were enriched in the pre-edited cells. Together, these data show that our approach of base editing phosphorylation sites in a pooled format can identify phosphorylated residues and putative kinases important for proper cell cycle proliferation or survival.

Coupling functional phosphosite screens with transcriptional reporters identifies novel regulators of NFAT transcriptional activity

To screen for phosphosites that are functionally linked to transcriptional outputs, we performed a “proteome-wide” base editor screen in activated TPR Jurkat cells, utilizing a GFP transcriptional reporter driven by the 4x ARRE2 sequence derived from the murine IL2 promoter³⁹. TPR Jurkat cells stably integrated with the sgRNA library described above were electroporated with ABE8e protein, stimulated with α-CD3/CD28 antibodies for 12 hours, and sorted for high and low GFP (NFAT activity) levels (Figure 4A). Genomic DNA was collected from the cells sorted into high and low bins, sgRNA identifiers were amplified by PCR, and were sequenced by next-generation sequencing (NGS). Total read-normalized, log transformed sgRNA counts were moderately to strongly correlated indicating sufficient data quality between quadruplicates, and the mean correlation between sgRNA counts in GFP high and low bins was strong (Extended Data Fig. 4A-B and Supp Table 4). We used MAGeCK¹³ to identify and rank which phosphosite mutations, the targets of the sgRNAs, regulate the NFAT transcriptional reporter. Phosphosites with multiple sgRNAs (Extended Data Fig. 4C) were combined in MAGeCK and the high and low GFP bins were compared. We identified 411 sgRNAs enriched in the GFP high bin, and 293 in the GFP low bin (Figure 4B and Supp Table 4). Rolling up our phosphosite level perturbations to the gene level (gene-centric), we performed pathway analyses, using multiple tools, to assess the fidelity of our approach. Enrichment analysis using MAGeCKFlute⁴⁰ identified TCR pathways as enriched in the GFP low (Figure 4C). g:Profiler⁴¹, a Gene Ontology-based analysis, also identified TCR pathway in the GFP low bin (Extended Data Fig. 4D). GSEA²⁵ identified the “TCR Calcium Pathway” signature in the GFP high bin (Supp Figure 4E), likely due, in part, to dephosphorylation of NFAT as a regulatory mechanism⁴². Edits containing bystander mutations next to the target phosphosite were not enriched in the GFP high and low bins compared to the library as a whole, suggesting they do not confer extra effects (Extended Data Fig. 4F).

Figure 4.

Proteome-wide base editing of phosphosites modulating NFAT transcriptional activity. A) Diagram of PTM-centric, proteome-wide base editing coupled to NFAT-GFP transcriptional reporter followed by next-generation sequencing (NGS). B) Volcano plot comparing phosphosite edits in GFP high (red) compared to GFP low (purple) bins as determined by MAGeCK. Statistical thresholds were also determined by MAGeCK¹³. C) MAGeCKFlute gene-centric pathway analysis of genes with mutated phosphosites enriched in the GFP low (purple) or GFP high (red) bins. Genes in the respective pathways and their fold change are shown. D) Kinase Library, site-centric enrichment analysis of phosphosite mutants enriched in the GFP low (purple) or GFP high (red) bins. Enrichment Values by MAGeCK¹³ and the one-sided Fisher’s exact text p value, after Benjamini-Hochberg, correction is show. E) PTM-set Enrichment Analysis (SEA) of phosphosites mutated by ABE8e protein and enriched in the GFP low (purple) or GFP high (red) bins. “iKiP” indicates the In Vitro Kinase-to-Phosphosite Database⁶⁷, “pKS” is kinase-substrate pairs from PhosphositePlus (PSP). MAPK14 is also known as P38A.

Utilizing PTM site-centric analyses, Kinase Library identified several kinases implicated in NFAT activity regulation, such as JNK⁴³, NLK⁴⁴, and CAMK2G⁴⁵, enriched in the GFP low bin (Figure 4D). The GFP high bin was enriched for the CDK9 motif, a kinase involved in general transcriptional regulation⁴⁶, as well as motifs of kinases known to be involved in T cell activation such as PAK2⁴⁷ and CDK5 ⁴⁸. CLK3, PAK4, SRPK1 and MYLK4 were also enriched in the GFP high bin but have poorly characterized roles in controlling NFAT transcriptional activity (Figure 4D). PTM-SEA²⁴ agreed with the Kinase Library results for PAK2 and CDK9 (Figure 4E). However, PTM-SEA also identified mutations of MAPK1, MTOR, and DYRK1A/2/3 substrates to be enriched in the GFP low bins, corroborating these kinases’ involvement in T cell activation (Figure 4E). As expected, these results demonstrate that phosphosite mutations can directly implicate their involvement in regulating NFAT transcriptional activity, recapitulating known signaling pathways that were constructed from various studies of general T cell activation. These results also strongly suggest that our approach of base editing phosphosites is not only capable of rediscovering crucial signaling molecules in T cell activation, but provides new insights by identifying novel, regulatory kinases and phosphorylation events.

We next asked how phosphopeptide abundance dynamics related to functional readouts in our NFAT-GFP screen. As phosphopeptide abundances could have increased and/or decreased in our time course data, we plotted the F statistic from the mass spectrometric analyses against the log2 fold change in the GFP high and low bins calculated by MAGeCK. We found that there was an overlap between the two datasets, where MAPK1 Y817C was amongst the largest changes in phosphopeptide levels and GFP expression (Extended Data Fig. 4G). We also tested whether phosphosites predicted to be functional by Ochoa et al. 2020⁴⁹ correlated with our empirical data for NFAT-GFP activity. Although there was a positive trend between our phenotypic screen data and predicted functional scores, the correlation was not statistically significant (Extended Data Fig. 4H-I). These results underscore the need for PTM-centric phenotypic screens.

PTM-centric base editor screening identifies the uncharacterized phosphorylation-mediated nuclear localization of PHLPP1

PHLPP1 is a protein phosphatase most-widely implicated in AKT signaling in cancer^50,51. In macrophages, PHLPP1 attenuates the JAK/STAT axis by dephosphorylating STAT1⁵². The post-translational mechanisms controlling PHLPP1 function, and its involvement in T cell biology⁵³, are poorly understood. We identified the mutation PHLPP1 S118P in our base editing screen to have a strong, negative impact on NFAT transcriptional activity, on par with MAPK1 Y187C (Figure 5A). We nucleofected ribonucleoproteins comprised of in vitro transcribed sgRNA coupled to ABE8e protein to validate our screen results. The intergenic mutation at the HEK3 site was used as an editing control. For positive effect controls, we introduced a terminating edit in the scaffolding protein LCP2 (SLP76), a critical molecule for proper T cell activation¹² or the MAPK1 Y187C mutation which prevented or altered T cell activation, respectively. After bulk editing was assessed by amplicon sequencing (x̅ = 70%, Extended Data Fig. 5), four to eight validated single cell clones were combined to avoid clone-specific effects. Editing at the top two predicted off-target sites for all clones was within the noise of the assay, about <3% editing. MAPK1 Y187C and PHLPP1 S118P both showed diminished NFAT transcriptional activity (Figure 5B). However, unlike MAPK1 Y187C, PHLPP1 S118P showed a small but statistically significant increase in CFP levels (NFκB activity reporter). To extend this analysis, we performed transcriptional profiling of clonal Jurkat T cells harboring the homozygous HEK3 intergenic edit, LCP2 terminating edit, the MAPK1 Y187C, or the PHLPP1 S118P mutations for zero and six hours post CD3/CD28 stimulation. LCP2-terminated cells showed no effect of transcriptional activation six hours after activation, whereas MAPK Y187C showed an intermediate pattern compared to the HEK3 controls (Figure 5C, Supp Table 5, and Supp Data File 2). The PHLPP1 mutant cells were similar in their transcription patterns compared to the HEK3 editing controls, though differences were apparent (Figure 5C). We highlighted the genes differentially expressed between HEK3, MAPK1 Y187C, and PHLPP1 S118P, and plotted them alongside the LCP2 terminating edit cells (Figure 5D). After k-means clustering and GO analysis of differentially expressed genes, clusters one, five, and six, which were higher in the PHLPP1 mutant compared to the HEK3 control identified multiple terms associated with NFκB signaling (NF-kappaB complex, TNFR signaling), corroborating our transcriptional reporter results (Figure 5B & D). The MAPK1 Y187C mutant cells were enriched for genes in sterol and isoprenoid biosynthesis (cluster 4 of Figure 5D). These results suggest different phosphorylation sites in the T cell activation pathway can regulate downstream gene expression in disparate ways.

Figure 5.

Phosphorylation-induced nuclear translocation of PHLPP1 promotes NFAT and represses NFκB transcriptional responses. A) Distribution of log2 fold changes of the ~11,000 sgRNAs inducing phosphorylation mutations in the NFAT (GFP) transcriptional activity screen (top, green) and non-targeting and intergenic controls (lower panel, gray) between GFP low and GFP high bins. PHLPP1 S118P and MAPK1 Y187C mutations are labeled for comparison. B) Validation of NFAT-activity screening hits using electroporation of in vitro transcribed sgRNAs coupled with ABE8e protein, followed by α-CD3/CD28 stimulation for 16 hours and analysis of GFP (NFAT) or CFP (NFκB) transcriptional activity reporters. Two sample T test p values compared to HEK3 relative color control are shown if > 0.001, n=3 activation replicates, where standard deviation is shown. C) Heatmap of differentially expressed genes between HEK3 intergenic mutant control, MAPK1 Y187C, or PHLPP1 S118P edited Jurkat cells activated for 0 or 6 hours as determined by single cell RNA sequencing. The LCP2 terminating edit is shown for comparison but was not used for statistical testing. (Supp Data File 2). D) K-means clustering and g:Profiler gene ontology analysis (all p < 0.05) of the differentially expressed gene clusters at 6 hours between HEK3 intergenic mutant control, MAPK1 Y187C, or PHLPP1 S118. The LCP2 terminating edit is shown for comparison but was not used for statistical testing. Cluster numbers count from left to right and are designated by color. E) Spinning disk confocal microscopy images showing the subcellular localization of PHLPP1 N-terminal extension (NTE) constructs. “NLS mut” refers to full mutation of the two nuclear localization sequences in PHLPP1’s NTE. The S118P, S118A and S118E mutations are also shown. F) Quantification of PHLPP1 NTE constructs and the percentage of α-HA signal in the nucleus. Unpaired, two sample T test p values are shown. Each data point represents an individual cell, where standard deviation is plotted.

S118 lies within the bipartite nuclear localization sequences (NLSs) at the N-terminus of PHLPP1⁵². To test the hypothesis that phosphorylation controls PHLPP1 subcellular localization, we ectopically expressed N-terminal extension (NTE) constructs with the wild type sequence, both halves of the NLS mutated, or S118P, the result of A-to-G editing. We also included the more archetypal amino acid substitution, S118A, which removes the phosphorylatable residue without potential structural changes introduced by the rotationally constrained amino acid proline, and the S118E mutation as a potential phosphomimetic. We found that both S118P and S118A reduced PHLPP1 NTE nuclear localization to comparable levels, but less severe than the full NLS mutant (Figure 5E-F). S118E showed a similar loss of nuclear localization compared to S118A or S118P, indicating in this case the glutamate may not accurately reflect a bona fide phosphorylation⁵⁴. These results suggest that phosphorylation of the NTE of PHLPP1 regulates nuclear localization. These data also suggest that serine to proline substitutions are reasonable proxies for loss of a phosphorylatable residue for screening purposes.

The T cell activation transcriptional program can be dissected through precise, base editor-mediated signaling modulation

Transcriptional profiling HEK3 (intergenic editing control), MAPK1 and PHLPP1 mutant Jurkat T cells showed more differentially expressed genes six hours post stimulus than in resting conditions (BH corrected p value <0.05) (Figure 5C), suggesting the mutated phosphosites indeed affect T cell activation-induced transcriptional responses. Inspection of the differentially expressed genes at six hours post T cell activation showed specific T cell-related genes are expressed at subtle but different levels between phosphosite-mutant genotypes (Figure 6A and Extended Data Fig. 6A). For example, BCL11A and THEMIS were activated to a higher extent in MAPK1 Y187C mutants compared to PHLPP1 S118P or control. In contrast, NR4A3, ZFP36L1, and IL21R were highest in the PHLPP1 S118P cells, whereas GZMA, TNFSF14, NFKBIA, and JUND were highest in the editing control cells (intergenic HEK3). Intracellular GZMB staining 24 hours after T cell activation corroborated the gene expression results, showing a loss of GZMB protein in MAPK1 Y187C but not PHLPP1 S118P cells (Figure 6B and Extended Data Fig. 6B). These results suggest that mutating different phosphosites in ostensibly the same signaling pathway can alter transcriptional responses, and may provide a means for fine-tuning gene expression.

Figure 6.

Dissecting T cell activation transcriptional responses. A) Select T cell-related differentially expressed genes at six hours post T cell activation between HEK3 (control), PHLPP1 S118P, and MAPK1 Y187C mutant cells. This data in log2 fold change space is available as Supp. Data file 2. B) Intracellular GZMB staining in HEK3 (control), PHLPP1 S118P, MAPK1 Y187C or LCP2 term edited mutant cells 24 hours post-T cell activation. Representative plot (left) and quantification of % GZMB⁺ cells, n = 6 activation replicates where standard deviation is shown.

Discussion

Linking specific signaling events to their downstream functions is a fundamental goal of biology. Choosing novel phosphosites for mechanistic follow-up studies is often resource intensive and laborious. We aimed to create a screening platform to functionally assess and prioritize individual phosphosites and how they contribute to specific phenotypes in high-throughput by integrating mass spectrometry-based phosphoproteomics with CRISPR-mediated base editor screens. We found that mutation of phosphosites via base editing in a pooled format can assess multiple functional readouts, in positive and negative directions, for proliferative or transcriptional phenotypes. This novel experimental and computational framework will greatly enable future studies, allowing the community to address the complexity of signaling pathways.

Our study focused on phosphosites empirically identified in a parallel experiment rather than from PTM repositories or databases, mitigating the introduction of indiscriminate coding mutations that can have PTM-independent effects^16,17,19,33. Moreover, as phosphoproteomic analyses of novel, unstudied systems continue to grow (i.e. patient-specific cancers, primary mouse or human immune cells, etc.) the bioinformatic tools needed to identify base editor-targetable phosphosites from empirical mass spectrometry data will become increasingly important (Figure 1D).

All gene-centric pathway analyses from genes with mutated phosphosites that altered NFAT activity revealed that the TCR pathway was enriched, indicating the results from our phosphosite base editor screen can recapitulate known aspects of well-studied pathways. Interestingly, gene-level GSEA analysis found “TCR Calcium Pathway” enriched in the GFP high bin. The phosphosite mutations driving this result were all in NFAT isoforms, or molecules known to regulate NFAT dephosphorylation: NFATC1/2/3, CABIN1, and RCAN1. This pathway regulates gene expression, cytokine production, and T cell activation through the NFAT signaling pathway by dephosphorylation of NFAT molecules, replicating the effect of Calcineurin (PPP3CC) in the translocation of NFAT and T cell activation^42,55. Site-centric pathway analyses of enriched phosphosite mutations revealed two major trends in the signatures identified in our Jurkat T cell activation model: TCR signaling and cell cycle. As Jurkat cells are rapidly dividing transformed cells²¹ it is not surprising that many of the phosphosites identified by mass spectrometry, then subsequently in the base editor screens, identified cell cycle genes and phosphorylation events. The DYRK family of kinases is a good example (Figure 3E and 4E). DYRKs are well known to control NFAT transcriptional activity^56,57, but also have roles in T cell proliferation⁵⁸. Our results suggest DYRKs have a stronger role in T cell activation compared to proliferation. We envision that applying base editor screens of PTM sites in primary immune cells will provide clearer connections within signaling-to-transcription networks.

Our PTM-centric base editor phenotypic screen identified a novel mechanism for PHLPP1 as a regulator of T cell activation-mediated gene expression. PHLPP1’s function has been primarily studied in cancer contexts^50,51, though its role in regulatory T cell development was identified through a genetic study⁵³. Our analyses originally identified it as a positive regulator of NFAT activity, and through transcriptional profiling we found that PHLPP1 can negatively regulate genes downstream of NFκB. This regulation is likely due to the phosphorylation-induced nuclear translocation of PHLPP1, though the precise substrates of dephosphorylation via PHLPP1 remain to be determined. It is worth noting that the phosphopeptide identifying PHLPP1 pS118 was induced at nine minutes 1.6-fold on average but was not statistically regulated in our phosphoproteomics data (Supp Table 1). Conversely, MAPK1 Y817C had both one of the strongest fold changes at the phosphopeptide abundance level and in the NFAT-GFP screen. The complimentary but orthogonal information between the phosphoproteomics and the base editor screen results is likely due to the specific nature of our chosen screening platform, an NFAT activity reporter. T cell activation is highly complex and not limited to transcriptional outputs, and thus, a substantial proportion of phosphorylation events may be completely unrelated to induction of NFAT activity. This underscores the need for functional screens of PTM sites as small changes to catalytic enzymes may have prominent downstream effects.

Gene expression profiling of various phosphosite-mutant Jurkats after T cell activation revealed subtle but different transcriptional responses. For example, the MAPK1 phosphomutant showed decreased expression of GZMB compared to the PHLPP1 mutant, at the protein and mRNA level (Figure 6), despite having similar effects on NFAT reporter activity (Figure 5). MAPK1 phosphomutant cells expressed less NR4A1 and NR4A3 than the PHLPP1 phosphomutant cells. GZMB is the major effector molecule of the cytotoxic program and NR4As have been shown to be drivers of T cell exhaustion⁵⁹. These results together lead to an intriguing possibility of fine-tuning expression levels of specific genes, through genetic or small molecule inhibitor manipulation of signaling pathways.

Proteome-wide, PTM-centric base editing coupled to phenotypic screens provides a powerful experimental framework to untangle the vast network of biochemical signaling reactions and how they lead to the control of specific cellular functions. This phosphosite-targeting base editing approach directly assesses the impact of a phosphosite mutation on a given phenotype rather than relying purely on evolutionary or structural conservation, and provides interpretable results with high confidence for further mechanistic studies^33,49,60,61. With currently available base editors it should be possible to functionally screen other PTMs including acetylation/ubiquitination (lysine), O-GlcNAcylation (Ser/Thr), cysteine¹⁸, or even specific proteolysis events (caspases), utilizing established infrastructure common to many research institutions. We envision this approach to be widely enabling to the cell biology community.

Limitations and future steps for PTM-centric base editor screens

We focused on ABE8e for our technology development because it can be readily expressed and purified from Escherichia coli, and its activity after electroporation into cells harboring sgRNA-expressing plasmids is highly efficacious (Figure 2C-D)^26,27. Also, as an A-to-G editor it can mutate tyrosine residues, a small but important fraction of total phosphosites. This property is missing from C-to-T editors. The amino acid substitutions made by current base editors to study phosphorylation are not archetypal (S/T to A; Y to F), threonines targeted by ABE8e being the exception (Thr to Ala). Over half of the serines mutated in our study were to proline, a limitation of current base editors. However, since several kinase families are proline directed and the majority of phosphosites are in flexible loops, this is likely to be problematic in only a subset of cases^62–64. Empirically, the vast majority of Pro substitutions in our data had no effect, arguing it is not inherently disruptive. In fact, our mutational analysis of the PHLPP1 phosphosite showed similar effects between the S118P and the archetypal S118A mutation used for proper validation, suggesting proline mutations are not invariably detrimental to screen for phosphosite function. Of course, it is prudent to validate the findings of a screen with orthogonal approaches. Prime editors, which can install specific, desired sequences, will likely become increasingly useful as they become adapted for genome-wide screens⁶⁵. They may also allow for installation of phosphomimetic substitutions, S/T/Y to D or E. However, carboxylic side chains can often not properly mimic a bona fide phosphorylated residue⁵⁴. Moreover, employing Cas molecules with less restrictive PAM sequences (NG rather than NGG), should double to triple the percentage of targetable phosphosites¹⁹.

Our nucleofection of purified ABE protein approach borrows from previous literature performing CRISPR-mediated gene knockouts in primary human myeloid and T cells, the latter of which can be performed at a genome-wide scale^12,66. As our phosphosite-targeting sgRNA library is much smaller than standard genome-wide libraries, our PTM-centric workflow should be readily adapted to other systems. Moreover, as mass spectrometry-based phosphoproteomics becomes more sensitive and robust, and unstudied systems such as patient-specific cancers or a variety of human or mouse immune cells are starting to be analyzed, the tools developed here should enable the community to study new signaling pathways in a more comprehensive and non-biased manner.

Methods

Cell Culture

Jurkat E6.1 and HEK293T cells were purchased through ATCC. Triple parameter reporter (TPR) Jurkat cells³² were purchased from Professor Peter Steinberger at the Medical University of Vienna. TPR Jurkat cells and E6.1 Jurkat cells were cultured and passaged in Roswell Park Memorial Institute medium 1640 (RPMI-1640) plus GlutaMax (Thermo Fisher Scientific) supplemented with 10% heat inactivated fetal bovine serum. Cells were passaged and maintained at cell densities between 1–5e5 cells per mL.

HEK293T cells were cultured and passaged in Dulbecco’s Modified Eagle Medium (DMEM) supplemented with 10% heat inactivated fetal bovine serum.

All cells were incubated and maintained at 37°C with 5% CO2.

Phosphoproteomic analysis

Jurkat E6.1 cells (ATCC) were activated in 96 well tissue culture plates for the stated times at 2e5 cells/mL. Plates were coated with 3.33 μg/mL α-CD3 (HITa3) and α-CD28 antibodies (Biolegend), in 100 μL PBS at 4°C overnight, followed by one cold PBS wash. To stop the reaction, the cells were transferred to ice cold PBS, washed twice, flash frozen, and stored at −80°C until processing. Phosphoproteomic analysis was performed as previously described²³. Data was analyzed using Spectrum Mill (Agilent and Broad Institute) for phosphopeptide identification and quantification. The TMT denominator for each sample was the mean of all TMT channels. After global median normalization and median absolute deviation scaling, a moderated F test (limma, R) was performed to identify “regulated” phosphopeptide levels across the time series. Statistically significant phosphopeptides (Benjamini-Hochberg adj. p value <0.05) were visualized using Morpheus (https://software.broadinstitute.org/morpheus/). The .json file associated with this manuscript, Supplemental Data File 1, can be used to explore these data in Morpheus.

ABE8e protein expression and purification

Recombinant ABE8e was expressed and purified as previously described²⁷. Briefly, ABE8e was expressed as an 8xHis tagged protein from a rhamnose-inducible promoter in BL21-Star DE3 cells with low RNase activity (Thermo Fisher Scientific). At an OD600 of ~0.8, the 2X terrific broth Escherichia coli cultures were cold shocked on ice for one hour, and induced with 0.8% final concentration of rhamnose. Roughly 24 hours later, cells were lysed via lysozyme and sonication, and ABE8e protein was purified on Ni-NTA resin. After imidazole elution, ABE8e protein was further purified using cation exchange. Fractions were monitored by UV and SDS-PAGE. ABE8e protein containing fractions were pooled, concentrated to ~90 pmol/μL using 100 MW cutoff filters, aliquoted, flash frozen, and stored at −80°C.

Base editing with arrayed lentivirus

Oligonucleotides containing the protospacer were chosen by hand⁶⁷ or from our original bioinformatic analysis, and ordered from IDT DNA technologies. CACCG was added to the forward oligo, where AAAC was added to the reverse complement of the protospacer sequence. A C on the 3’ end of the reverse complement oligo was also appended. pRDA118 (Addgene Plasmid # 133459) was digested with BsbmI_V2 (NEB) and FastAP (Thermo) for the last five minutes, followed by gel purification. Oligos were mixed 1:1 at 100 μM, phosphorylated via PNK (NEB) and annealed after a 5 minutes 95°C step at 5°C per five minutes until 25°C. Annealed oligos were diluted 1:200 and 1 μL was mixed with 25–50 ng of digested backbone. T4 ligation (NEB) was performed for 20 minutes at 37°C and the 5–10 μL ligation reaction was transformed into Stbl3 E. coli, made in house. Colonies were verified via Sanger sequencing. One μg of sgRNA plasmids were transfected with Lipofectamine 3000 (Invitrogen) 1 μg of pPAX2 and 0.1 μg VSVG into HEK293Ts seeded the night before at 2e5 cells per 6 well. After three hours the 2.5 mL media was replaced with 5 mLs media supplemented with 1% BSA. Supernatants were collected after 3 days, filtered to remove HEK293T cells, concentrated 10-fold with Lenti-X precipitation solution (Alstem), and aliquots were flash frozen and stored at −80°C.

Wild-type or TPR Jurkat cells were spinfected for two hours at 666 x g. Two days later, 2 μg/mL puromycin was added until the non-transduced cells were all dead (2–4 days). To test base editor delivery molecules, only the HEK3 site-targeting sgRNA was used, and electroporation was performed using Lonza Nucleofection with the SE or P3 nucleofection solution. pCMV-BE4max was used for plasmid DNA. Base editor mRNAs were designed⁶⁸ generated by in vitro transcription using the HiScribe T7 High-Yield RNA synthesis kit (NEB Cat No. E2040S)⁶⁹. NEBnext polymerase was used to PCR-amplify template plasmids and install a functional T7 promoter and a 120 nucleotide polyadenine tail. Transcription reactions were set up with complete substitution of uracil by N1-methylpseudouridine (Trilink BioTechnologies Cat No. N-1080) and co-transcriptional 5’ capping with the CleanCap AG analog (Trilink BioTechnologies Cat No. N-7113) to generate a 5’ Cap1 structure. mRNAs were purified using ethanol precipitation according to kit instructions, dissolved in nuclease-free water, and normalized to a concentration of 2 micrograms per microliter using Nanodrop RNA quantification of diluted test samples.

Base editing with in vitro transcribed guides and ABE8e protein

sgRNAs were transcribed in vitro using the EnGen sgRNA Synthesis kit (NEB) with oligonucleotides containing the protospacers (Supp. Table 6). The sgRNAs were then purified (NEB Monarch RNA Cleanup kit) and quantified by nanodrop.

3 μg of purified sgRNA were then mixed with 1 μL of 90 μM ABE8e protein in Lonza P3 electroporation buffer in a total reaction volume of 10 μL and incubated at room temperature to allow ribonucleoprotein complex to form.

TPR Jurkat cells were collected, spun, washed twice with 37°C PBS, and 2e5 cells were resuspended per cuvette well in Lonza electroporation buffer P3 with the ABE8e and sgRNA ribonucleoprotein complex to a total reaction volume of 20 μL. Electroporation and cell recovery was performed according to the manufacturer’s instructions. Two days post nucleofection, gDNA of the cells was extracted, and base editing efficiency of each population was determined through PCR amplification of the edited genomic region of interest and quantified through Sanger sequencing and EditR software analysis²⁸. To produce purely edited cell populations, single cell clones were isolated from the corresponding bulk edited cell populations by diluting 0.8 cells per well in 96-well plates. Once isolated, these cells were grown until confluence, then individually genotyped through extraction of its gDNA, PCR amplification of the edited genomic region of interest, and Sanger sequencing via EditR²⁸. Once single cell clones were validated by Sanger sequencing, 4–8 single cell clones were mixed together to avoid clone-specific effects.

CD69 staining

Cells activated as described above were washed once with cold Cell wash buffer (Biolegend) and incubated with 0.5 uL α-CD69-APC antibodies (Biolegend) at 4°C, wrapped in foil, for 20 minutes. Cells were washed twice with cold PBS and stored on ice prior to flow cytometry.

Phosphosite library design and construction

All phosphorylation sites determined through the accompanying phosphoproteomic analysis were filtered for phosphosites that were localized in the peptide with a confidence of 90% or greater. For all genes with phosphosites, we used an in-house base editor design tool to first design all possible guides targeting these genes. All adenines in the window of 4–8 of the sgRNA (where 1 is the most PAM-distal position, and positions 21–23 are the PAM) were considered to be edited. sgRNAs with cloning sites, poly Ts and greater than five perfect matches in the genome were excluded. sgRNAs targeting the phosphosite of interest were then picked. sgRNAs predicted to make silent edits at phosphosites but also predicted to have bystander edits in the window of 4–8 were excluded. We then included 250 non-targeting controls, 250 intergenic controls, 250 controls targeting splice sites of essential genes and 250 controls targeting splice sites of TCR genes. The code is available at https://github.com/mhegde/base-editor-design-tool.

Proteome-wide base editor screen

TPR Jurkat cells were spinfected in 4 µg/mL polybrene at 30°C for 45 minutes at 666 x g at a multiplicity of infection of 0.3, assuming 30% transduction efficiency, and maintained a 500x library coverage. Transduction quadruplicates were used for downstream replicates. Puromycin was added to 2 μg/mL and cells were selected for stable integrants for seven days. Three days after removal of puromycin, an aliquot of cells corresponding to 500x library coverage was saved as the pre-ABE8e inputs. Cells were washed with 37°C PBS twice, and resuspended in SE nucleofection reagent (Lonza) with 94 pmoles of ABE8e protein per well. Each well was 2e5 library-containing TPR cells in 19 µL of SE + 1 µL ABE8e protein, and all 16 wells were used simultaneously for a library coverage of 250x. Electroporation and cell recovery was performed according to the manufacturer’s instructions. Six days after ABE8e protein introduction, another 500x aliquot of cells was frozen for the post-ABE8e sample.

14 days after ABE8e protein introduction, the mutant library TPR Jurkat pool was washed in room temp PBS, and activated as described above. 2000x library coverage of TPR cells was prepared for FACS. FACS was performed on a Bigfoot Spectral Cell Sorter (ThermoFisher) and roughly 1e6 cells were collected per bin. The top one, and bottom two 12.5% bins were sorted. The two bottom bins (“bottom” and “low”) were sorted to avoid unactivated cells, which always was about 15% of all cells. Only the second lowest bin (“low”) was used for downstream analyses due to the superior performance of controls, and to ensure cells were activated.

Collected cells were pelleted and stored at −80°C until further processing. Genomic DNA (gDNA) was isolated and PCR-amplified for barcode abundance determination as previously described¹⁶. Standard Illumina adapters were added and stagger regions were introduced for base diversity, according to the protocol “sgRNA/shRNA/ORF PCR for Illumina Sequencing” (The Broad Institute GPP). Libraries were sequenced at LJI a depth of 10e6 reads or more.

Differential sgRNA abundance analyses via MAGeCK and MAGeCKFlute Programs

To identify the functional phosphorylation sites through mutational analyses, we analyzed raw reads of GFPhigh and GFPlow samples using the MAGeCK program (v0.5.9.5), used for analyzing CRISPR screens¹³. To summarize the results of MAGeCK’s phosphosite and sgRNA data, we used MAGeCKFlute program (version 2.4.0)⁴⁰. Intergenic and non-targeting controls were used for normalization and size factor estimation. Before conducting the pathway enrichment and gene-centric GSEA analysis with MAGeCKFlute, we converted the phosphorylated and mutated gene products to their corresponding gene symbols. For site-centric analyses (PTM-SEA, Kinase Library), we used the MAGeCK output for ssGSEA2.0 R package with default settings²⁴. PTM-SEA included iKiP data⁶⁷ but excluded the LINCS P100 terms for clarity. To investigate the putative kinases responsible for phosphorylating phosphosites abundant in post-base editing and NFAT signaling, we employed The Kinase Library Enrichment analysis, which offers an atlas of primary sequence substrate preferences for the human serine/threonine kinome, was used on the PhosphositePlus.org website³⁸.

Data visualization is done using R 4.3.0, Graphpad Prism 10, Protigy (v1.1.7), iDEP 0.96.

Dataset comparisons

To compare the mass spectrometry-based phosphoproteomic data to the phenotypic screen results, we used the F statistic calculated during the moderated F-test as our variable. We reasoned that since we had multiple time points, where multiple pairwise comparisons could be made, the F statistic would capture the magnitude and reproducibility of a phosphopeptide’s abundance over time. Directionality is lost, though directionality may differ between any two different time point comparisons.

To compare predicted functional scores, we downloaded the Supplemental Tables from Ochoa et al. 2020⁴⁹ (Supp Table 3) and compared the column “functional_score” to the log2 fold change of the base editor screens calculated by MAGeCK.

PHLPP1 localization

Plasmids encoding the N-terminal extension (NTE) of PHLPP1 were HA tagged⁵², and the appropriate codons were mutated using site-directed mutagenesis (Agilent). HeLa cells were transiently transfected with WT, NLS mutant, S118P, S118A, or S118E PHLPP1 NTEs, and after two days were fixed, and stained with anti-HA antibody(1:500 dilution) (Cell Signaling, 3742), Alexa Fluor-647-phalloidin (Invitrogen, A22285), and DAPI stains. Images were acquired using spinning disk confocal microscopy. Quantification was performed using individual cells and were statistically tested using a one-way ANOVA with Tukey’s multiple test corrections.

Transcriptional profiling of phosphosite-mutant T cell lines

The various phosphosite mutants were introduced into TPR Jurkat cells via in vitro transcribed sgRNAs described above. After single cell cloning 4–8 single cell clones were mixed together to avoid clone-specific effects. Cells were incubated and activated with α-CD3/CD28 agonist antibodies as described above, except that the antibody concentrations were 1 μg/mL. Cells were activated for 0 and six hours.

Cells were stained with BioLegend TotalSeq-C Human Universal Cocktail and anti-human hashtag antibodies according to NYCG CITEseq protocols [http://cite-seq.com/] and processed using 10X Genomics 5’ HT with Feature Barcode assay according to protocol, with the addition of selective transcript removal using Jumpcode Genomics CRISPRclean Single Cell Boost kit.

Briefly, 100K cells from each phosphosite-mutant line per time point were blocked with BioLegend Human TruStain FcX Fc receptor blocking solution. Cells were then stained with BioLegend TotalSeq-C Human Universal Cocktail resuspended in BioLegend Cell Staining Buffer (CSB) at a concentration of 1 vial per 500K cells and anti-human hashtag antibodies at a concentration of 0.75 μg/1e6 cells for 30 minutes at 4°C. Cells were then washed 3x in CSB. Cell viability and concentration was assessed using the Moxi Go II and Moxi Cyte Viability Reagent containing propidium iodide (PI). Cells were pelleted and resuspended in 0.04% BSA in PBS for a final concentration of 1300–1600 cells/μl and processed using 10X Genomics Next GEM Single Cell 5’ HT v2 assay. Gene Expression (GEX) and Cell Surface Protein (CSP) libraries were constructed according to protocol (CG000424 Rev D) with the following deviation: Post-ligation product of the GEX library was subjected to Jumpcode Genomics CRISPRclean Single Cell Boost kit following protocol. Libraries were sequenced at a targeted depth of 50K reads/cell for GEX and 5k reads/cell for CSP on an Illumina NovaSeq 6000 and an Element Aviti. Cellranger count v7.1.0 was used to generate cloupe files for analysis.

Data was analyzed using the Loupe Browser (10X Genomics). For Figure 5C, HEK3, MAPK1, LCP2, and PHLPP1 phosphosite-mutant cells were selected for zero and six hours post-activation, and differential gene expression (log2 fold change) was determined using local (sample specific) expression. Genes with a p value less or equal to 0.1 were determined to be regulated, as recommended by the software. For Figure 5D, differentially expressed genes were determined using only HEK3, MAPK1, and PHLPP1 cells, though LCP2 terminating edit cells’ log2 fold change values were included in the plot for context. For Figure 6A, genes were selected from the plot in 5D through their known involvement in T cell signaling. The .json file associated with this analysis, Supplemental Data File 2, can be used to explore these data in Morpheus.

Granzyme B staining

3.5e5 cells were incubated with fixable viability dye prior to fixation and permeabilization using Ghost Dye UV 450 (Cytek) at a 1:100 dilution, in a total volume of 50μL, for 15 minutes in PBS containing 2% FBS (PBS 2%) at 4°C and then washed once in the same medium at 500g for 4 min. Cells were fixed in 50 μL of BD Cytofix during 20 min at 4°C and then washed in PBS 2% at 500g for 4 min. After, cells were fixed and permeabilized using the eBioscience Foxp3 / Transcription Factor Staining Buffer Set (Thermo Fisher) for 40 minutes at 4°C and then washed in perm/wash buffer at 500g for 4 min. Expression of granzyme b was evaluated using the Granzyme B- A647 monoclonal antibody (clone GB11, BD biosciences) at a 1:100 dilution, in a total volume of 50μL. Cells were incubated 1 hour at RT followed by a 4°C incubation overnight. Cells were washed in perm/wash at 500g for 4 min and then resuspended for analysis. Analyses were performed on LSRII cytometer (BD Biosciences). 10,000 events were recorded and data analyses were performed in FlowJo software (Tree Star, Ashland, OR).

Extended Data

Extended Data Fig. 1. Extended Data Figure 1 associated with Figure 1 Kennedy et al.

Phosphoproteomics quality controls. A) Principal component analysis of all phosphoproteomics samples prior to differential expression analysis. B) Multi-scatter plot comparing all samples to each other pairwise. Pearson’s r is shown. Colors of samples are the same as in Extended Fig. 1A.

Extended Data Fig. 2. Extended Data Figure 2 associated with Figure 2 Kennedy et al.

Gating strategy for CD69 staining analyzed by flow cytometry.

Extended Data Fig. 3. Extended Data Figure 3 associated with Figure 3 Kennedy et al.

Gene ontology analysis of the genes targeted by the sgRNAs depicted in Figure 3D. Colors coordinate with Figure 3D. Dotted line is the hypergeometric distribution test FDR threshold.

Extended Data Fig. 4. Extended Data Figure 4 associated with Figure 4 Kennedy et al.

Quality control and characterization of phosphosite base editing coupled to NFAT activity reporters. A) Pairwise Spearman correlations between all normalized log transformed read counts across replicates and experimental conditions. 0.4 is the lower limit cut off in black. B) Mean (across replicates) sgRNA counts for individual sgRNAs prior to collapsing redundant phosphosite targets in the GFP high and low bins. Regression line is shown. C) Percentage of phosphosite targets with one or more protospacer sequences. D) g:Profiler (gene-centric) analysis of genes with phosphosite mutations enriched in the GFP low or GFP high bins. For the x-axis the normalized enrichment score (NES) was multiplied by the -log10 FDR. E) GSEA (gene-centric) analysis of gene sets enriched in the GFP high bin. TCR Calcium Pathway is bolded. F) Proportion of phosphosite targets that contain a putative bystander edit in the library as a whole and in the sorted GFP bins. Student’s two sample T test P value is shown. G) Scatterplot comparing the F statistic from the phosphoproteomic analysis, a proxy for magnitude and reproducibility of abundance changes across the four time points, and the log2 fold change GFPhigh/low bins calculated by MAGeCK. Horizontal red dashed line delineates nominal p value of < 0.05 from the moderated F test of the phosphoproteomics data. H) Scatterplot comparing the log2 fold change GFPhigh/low bins calculated by MAGeCK to the predicted functional score from the machine learning analysis in Ochoa et al⁴⁹. Inset shows the full data structure while the scatter plot is a zoom of points above a predicted functional score of 0.5. Horizontal red dashed line delineates a score threshold determined in Ochoa et al⁴⁹. I) Distribution of predicted functional scores from Ochoa et al⁴⁹. for all data points in the GFP screen, the phosphosite mutants that increased (“up” in red) or decreased GFP levels (“down” in purple). P values for comparison to the whole data set are shown. Data points represent the mean log2 FC (GFPhigh/GFP low) of four transduction replicates. P values for an ANOVA test followed by uncorrected Fisher’s least significant difference for multiple comparisons.

Extended Data Fig. 5. Extended Data Figure 5 associated with Figure 5, Kennedy et al.

EditR software analysis²⁸ plots outlining bystander base editing levels for PHLPP1 S118P and MAPK1 Y187C prior to single cell clone isolation.

Extended Data Fig. 6. Extended Data Figure 6 associated with Figure 6, Kennedy et al.

A) Log2 fold change of select T cell genes differentially expressed between PHLPP1 S118P and MAPK1 Y187C mutant cells, compared to HEK3 control cells. B) Gating strategy for intracellular GZMB staining and analysis by flow cytometry.

Data Availability

Raw mass spectrometry data and metadata can be accessed at ftp://MSV000092965@massive.ucsd.edu. Raw RNA sequencing data can be accessed at GEO accession ID GSE244164.